SEKE 2012 Proceedings - Knowledge Systems Institute

[29]. In their study they have presented empirical evaluations
showing that SLOC is a direct measure of complexity, the only
exception being header files which showed a low correlation
with the McCabe’s cyclomatic complexity measures. Research
by Graylin et al. presented evidence that SLOC and cyclomatic
complexity have a stable practically perfect linear relationship
that holds across programmers, languages, code paradig ms
(procedural versus object-oriented), and software processes
[30]. Linear models have been developed relating SLOC and
cyclomatic xomplexity. Buse et al [31], for example, presented
a study where they show a high direct correlation between the
SLOC and the structural complexity of the code. A study by
Gaffney [32] reported that the number of faults in a software
component is directly correlated to source lines of code
(SLOC). Krishnan et al. [33] also reported an empirical study
that showed a direct correlation between SLOC and the number
of defects in software components.
Based on these studies, we decided to compare the one-use
and reusable components in our study based on the size (in
SLOC), effort (man-hours) and productivity (in SLOC/hr).
V. RESULTS AND ANALYSIS
A. Reuse Design Principles
Table II shows the summary of the usage of reuse design
principles by the subjects. The mean number of principl es
used by a subject was 5.4. A well-defined interface (#1) was
the most highly ranked principle and was used for about half
of the reusable components. Link to documentation was
ranked 2 and used in about 42% of the reusable components.
Documentation has always been recommended and widely
used in the programming world. Clarity and understandability
of the code ranked next. This principle allows the users of the
component a better and easier way of comprehending the code
for future use. The next three principles were generality,
separate concepts from contents and commonality and
variability.
B. SLOC, Effort, Productivity and Parameters
The source lines of code for the one-use (N=107) and
reusable components (N=101 ) were measured using the
sloccount tool [http://www.dwheeler.com/sloccount/]. The
notched box plots of the SLOC measured for the one-use and
reusable components are shown in Fig. 2.
SLOC ReuseDiff = (SLOC Reuse – SLOC one-use ) (5)
The effort taken in terms of time (hours) and the number of
parameters are shown in Fig. 3 and 4 respectively. Fig. 5
compares the productivity (in terms of SLOC/ho ur) of the
developers for one-use vs. reuse components. SLOC ReuseDiff is
considered for the productivity of reusable components
because the reusable components studied in this paper were
not built from scratch; instead , they were reengineered by
modifying the one-use components.
SLOC/hr ReuseDiff = (SLOC Reuse – SLOC one-use ) / Effort Reuse (6)
TABLE II. Ranking of reuse design principles used
Rank# Reuse Design Principle Count#
1 well defined interface 56
2 link to documentation 43
3 clear and understandable 42
4 generality 41
5 separate concept from contents 40
6 commonality and variability 31
7 linking of test to code 24
8 encapsulation 23
9 one component use many 21
10 composition 19
11 variability mechanism 13
12 parameterization 12
13 genericity 11
14 optimization 9
15 restrictiveness 7
16 modification 3
17 isolate context and policy 1
18 abstraction 1
19 self-documenting code 1
Understanding and interpreting box plots can be found in
[34]. If the notches of boxpl ots of different groups overlap,
then there is no significant difference between the groups and if
they do not overlap, there is significant difference between the
groups.
The median of SLOC significantly increased for the
reusable components to 92 lines of code as compared to 51 for
the one-use components, an increase of 80%. The aver age
SLOC was 62 and 110 for the one-use components and the
reusable components respectively. The notches of the two box
plots do not overlap and this indicates a statistically significant
difference between the sizes of the tw o components. This
increase is due to inco rporating more functi onality in the
reusable components. The boxplots in Fig. 2 also shows that
there is much more variability in the SLOC measure for the
reusable components. This may be because the reusable
components have more functionalities and those functionalities
vary from subject to subject based on the understanding of the
reuse design principles; while for the one-use components the
subjects may have had a more similar understanding of the
functionality. From Fig. 3, the median of the time taken to
implement the components was 5.0 hours and 8.0 hours
respectively for the one-use and reusable components. Average
time taken was 3.6 and 6.5 hours for one-use components and
reusable components respectively. The notched areas of the
box plots overlap for the two and this indicates no significant
difference. As was the case for SLOC, the variability is higher
for the reusable components. The inter quartile range for the
one-use and reusable components are 3.0 hours and 5.5 hours
respectively while the standard deviations are 3.0 hours and 6.8
hours respectively.
197